And I didn't mean to ridicule Mark about suggesting too small of a filter
for his goals.  I just wanted to point out the wishful thinking for
something impossible.

Canon already has a couple of lenses with filter threads that are too small
already.  The EF 135mm f/2.8 SF lens has only a 52mm thread with a recessed
first element.  The lens should have a 48.2mm aperture however 52mm filters
have a clear diameter of only 47mm.  Although the filter doesn't cause
vignetting on a Rebel XT due the reduced sensor size, it does darken the
meter readings by 1/3 stop by putting on a clear filter that doesn't cause
any noticeable loss on other lenses.  The loss due to the diameter is less
than 1/3 stop but adding the loss of light from reflections and absorption,
it comes to 1/3 stop so it is noticable.  With other lenses the losses are
small enough for the meter to not notice this filter.

The EF 200mm f/2.8L II USM has a larger problem.  I.E. a 71.4mm aperture
with only a 72mm filter thread.  The inside diameter of a 72 x 0.75mm thread
is already only 71.25mm.  It is slightly too small already and even worse
when you consider that the light cone is converging into the lens.  There
are rays being clipped for sure.  My 72mm filters measure 67mm of clear
glass, so they are way too small.  A 67mm clear aperture reduces the lens to
at best f/2.99, a full sixth of a stop lost not including the cone angle and
filter factor.  The real losses are larger, probably more than 1/3 stop.
Usually the entrance pupil is actually smaller than the front element, but
clearly clear aperture of the front element must be at least as big as the
entrance pupil or else it becomes another iris limiting the light input.
Why didn't Canon use a 77mm filter thread?  77mmm filters have a clear area
of 72mmm so they would work wit less loss.

If anyone has a EF 200mm f:2.8L USM could you try metering the same scene
with and without a 72mm clear UV filter?

Probably not.  But this part of the thread was a conversation between myself
and "Rita."  Not between myself and Canon.  Also, it might be indicative of
what it would take for someone, such as myself, to upgrade from the present
16-35 to the new one, if they already had the present iteration, which I do.